Five AI Trends To Monitor In 2025

Fene Osakwe is one of Africa's most sought-after cybersecurity advisors, best-selling author, global speaker and cyber mentor.

One of the trends that shaped 2024 was artificial intelligence (AI). Many technology conferences I attended were eager for submissions related to AI, and TV interviews on technology trends consistently sought insights on this topic. A McKinsey survey shows that AI adoption increased from 50% to 72%. Companies are actively exploring ways to leverage AI to improve their business operations.

The landscape in 2025 is expected to be similar as we continue to navigate the potential of AI and examine how we can maximize its benefits.

The purpose of this article is to communicate primarily to a non-technical audience five AI trends to watch for in 2025 and how these trends can be leveraged to enhance various aspects of your business.

Agentic AI

In simple terms, agentic AI refers to systems that can reason, plan and act autonomously. Unlike generative AI, which many of us are familiar with, agentic AI can carry out complex sequences of activities and take corresponding actions. This includes tasks such as searching a database, retrieving results and triggering workflows without human intervention.

Organizations may be inclined to explore this capability further by 2025. For instance, in IT operations, predicting when a disaster recovery plan may need to be invoked and initiating workflow actions during an incident could prove to be extremely valuable.

Additionally, companies may leverage agentic AI for climate predictions and adapt supply chains accordingly. This article by Forbes Contributor Bernard Marr highlights some other benefits of agentic AI for businesses.

Inference Time And Large Language Models

According to IBM, "Large language models (LLMs) are a category of foundation models trained on immense amounts of data making them capable of understanding and generating natural language and other types of content to perform a wide range of tasks."

Inference time refers to the duration it takes these LLMs to take in a user query, process that query against their training data, perform reasoning and provide an output.

Currently, the challenge with inference time is its variability; reasoning may also deteriorate after a certain period. Issues can arise in which the model loses track of the task, starts repeating steps unnecessarily or even begins to output gibberish.

A report from Reuters highlights this issue and indicates that companies like OpenAI are aware of it. By 2025, I believe there will be an innovation to improve inference time, which will ultimately support the development of more intelligent AI agents.

Improved Use Cases

This article by Forbes Contributor Sol Rashidi highlights the top AI use cases in 2024. A closer examination of the themes among the top 10 reveals a significant increase in AI applications for content creation, such as CV reviews, emails, blog writing and grammar correction. These advancements are largely driven by generative AI.

As we enter the era of agentic AI, we can expect more use cases to emerge within business operations and technology. For instance, in cybersecurity, AI is being leveraged to respond in real time to zero-day attacks, which do not have signatures. As the trends mentioned earlier continue to grow, opportunities for businesses will expand, leading to improved use cases.

Smaller Models

One of the challenges with large language models is the immense computing power required to operate them. According to Spheron, massive models such as GPT-3, which has 175 billion parameters, require substantial memory of about 350 GB.

When you multiply this by the number of users in an average company, it becomes clear that some organizations may struggle to afford such high computing costs. One trend I predict for 2025 is the development of scaled-down versions of these models, which would make computing investments more affordable for individuals and organizations.

Privacy Concerns

In 2024, we saw the introduction of various regulations, including the PIPL in China and the EU AI Act, along with several other countries and regions beginning to establish frameworks, laws and best practices to ensure that the rights of citizens, employees and organizations are protected from AI systems.

By 2025, we can expect increased scrutiny regarding the ethics of AI. Several important questions need to be addressed: How do we audit the training data? How can we ensure that the training data has not been compromised?

I was part of a team that tested a model designed to help human resources quickly scan through thousands of applications and identify the best candidates. We observed that a particular race had been completely excluded from the output. This raises questions about unconscious bias, which the creators of these models and agents must address.

I believe that as we move into 2025 and 2026, a balance between innovation, governance and ethics will be essential to get the best from AI.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?